Model-Based Probabilistic Reasoning for Electronics Troubleshooting

1N-ATE is an on-going project aimed at developing expert consultant systems for guiding a novice technician through each step of an electronics troubleshooting session. One goal of the project is to automatically produce, given a set of initial symptoms, a binary (pass/fail) decision tree of testpoints to be checked by the technician. This paper discusses our initial approach using a modified game tree search technique, the gamma miniaverage method. One of the parameters which guides this search technique the cost of each test is stored a priori The two other param­ eters that guide it the conditional probability of test out­ comes and the proximity to a solution are provided by a dynamic model of an expert troubleshooter's beliefs about what in the device is good and what is bad. This model of beliefs is updated using probabilistic "tp.st-resuLt ► plausible-consequences" rules These rules are either pro­ vided by an expert technician, or approximated by a model-guided Rule Generator The model that guides the generation of rules is a simple block diagram of the Unit Under Test (UUT) augmented with component failure rates.